FPGA Implementation of a Pipelined Gaussian Calculation for HMM-Based Large Vocabulary Speech Recognition

A scalable large vocabulary, speaker independent speech recognition system is being developed using Hidden Markov Models (HMMs) for acoustic modeling and a Weighted Finite State Transducer (WFST) to compile sentence, word, and phoneme models. The system comprises a software backend search and an FPGA-based Gaussian calculation which are covered here. In this paper, we present an efficient pipelined design implemented both as an embedded peripheral and as a scalable, parallel hardware accelerator. Both architectures have been implemented on an Alpha Data XRC-5T1, reconfigurable computer housing a Virtex 5 SX95T FPGA. The core has been tested and is capable of calculating a full set of Gaussian results from 3825 acoustic models in 9.03 ms which coupled with a backend search of 5000 words has provided an accuracy of over 80%. Parallel implementations have been designed with up to 32 cores and have been successfully implemented with a clock frequency of 133?MHz.

Gaussian Approximation for Tracking Occluding and Interacting Targets

In this paper, we show how interacting and occluding targets can be tackled successfully within a Gaussian approximation. For that purpose, we develop a general expansion of the mean and covariance of the posterior and we consider a first order approximation of it. The proposed method differs from EKF in that neither a non-linear dynamical model nor a non-linear measurement vector to state relation have to be defined, so it works with any kind of interaction potential and likelihood. The approach has been tested on three sequences (10400, 2500, and 400 frames each one). The results show that our approach helps to reduce the number of failures without increasing too much the computation time with respect to methods that do not take into account target interactions.

A Kernel Interleaved Scheduling Method for Streaming Applications on Soft-core Vector Processors

Massively parallel networks of highly efficient, high performance Single Instruction Multiple Data (SIMD) processors have been shown to enable FPGA-based implementation of real-time signal processing applications with performance and
cost comparable to dedicated hardware architectures. This is achieved by exploiting simple datapath units with deep processing pipelines. However, these architectures are highly susceptible to pipeline bubbles resulting from data and control hazards; the only way to mitigate against these is manual interleaving of
application tasks on each datapath, since no suitable automated interleaving approach exists. In this paper we describe a new automated integrated mapping/scheduling approach to map algorithm tasks to processors and a new low-complexity list scheduling technique to generate the interleaved schedules. When applied to a spatial Fixed-Complexity Sphere Decoding (FSD) detector
for next-generation Multiple-Input Multiple-Output (MIMO) systems, the resulting schedules achieve real-time performance for IEEE 802.11n systems on a network of 16-way SIMD processors on FPGA, enable better performance/complexity balance than current approaches and produce results comparable to handcrafted implementations.

Macroscopic limit of time-dependent density-functional theory for adiabatic local approximations of the exchange-correlation kernel

Time-dependent density-functional theory is a rather accurate and efficient way to compute electronic excitations for finite systems. However, in the macroscopic limit (systems of increasing size), for the usual adiabatic random-phase, local-density, or generalized-gradient approximations, one recovers the Kohn-Sham independent-particle picture, and thus the incorrect band gap. To clarify this trend, we investigate the macroscopic limit of the exchange-correlation kernel in such approximations by means of an algebraical analysis complemented with numerical studies of a one-dimensional tight-binding model. We link the failure to shift the Kohn-Sham spectrum of these approximate kernels to the fact that the corresponding operators in the transition space act only on a finite subspace.

Gaussian mixture background modelling optimisation for micro-controllers

This paper proposes an optimisation of the adaptive Gaussian mixture background model that allows the deployment of the method on processors with low memory capacity. The effect of the granularity of the Gaussian mean-value and variance in an integer-based implementation is investigated and novel updating rules of the mixture weights are described. Based on the proposed framework, an implementation for a very low power consumption micro-controller is presented. Results show that the proposed method operates in real time on the micro-controller and has similar performance to the original model. © 2012 Springer-Verlag.

Gaussian pulse mixing and scattering by finite nonlinear layered structures

Multiple Gaussian pulse interactions and scattering in the nonlinear layered dielectric structures have been examined. The Gaussian pulses with different centre frequencies and lengths are incident at oblique angles on the finite stack of nonlinear dielectric layers. The properties of the reflected and refracted waveforms and the effects of the structure and the incident pulses' parameters on the mixing process are discussed. It is shown that the efficiency of forward emission at the combinatorial frequency can be considerably increased when the wavelengths of interacting pulses are close to the edges of electromagnetic bandgap. © 2012 IEEE.

GAUSSIAN PULSE SCATTERING BY NONLINEAR THUE-MORSE QUASIPERIODIC MULTILAYERS

The pulse mixing and scattering by finite nonlinear Thue-Morse quasi-periodic dielectric multilayered structure illuminated by two Gaussian pulses with different centre frequencies and lengths are investigated. The three-wave mixing technique is applied to study the nonlinear processes. The properties of the scattered waveforms and the effects of the structure and the incident pulses' parameters on the mixing process are discussed.

Gaussian pulse mixing by nonlinear photonic crystals

The properties of mixing and scattering of two non-collinear Gaussian pulses with different centre frequencies and lengths, incident on the finite nonlinear periodic layered dielectric structures, have been analysed. It is shown that at the backward emission grows with the number of layers and can reach the level of the forward emission in the direction of combinatorial frequency scattering.

Inverse Gaussian Modeling of Turbulence-Induced Fading in Free-Space Optical Systems

We propose the inverse Gaussian distribution, as a less complex alternative to the classical log-normal model, to describe turbulence-induced fading in free-space optical (FSO) systems operating in weak turbulence conditions and/or in the presence of aperture averaging effects. By conducting goodness of fit tests, we define the range of values of the scintillation index for various multiple-input multiple-output (MIMO) FSO configurations, where the two distributions approximate each other with a certain significance level. Furthermore, the bit error rate performance of two typical MIMO FSO systems is investigated over the new turbulence model; an intensity-modulation/direct detection MIMO FSO system with Q-ary pulse position modulation that employs repetition coding at the transmitter and equal gain combining at the receiver, and a heterodyne MIMO FSO system with differential phase-shift keying and maximal ratio combining at the receiver. Finally, numerical results are presented that validate the theoretical analysis and provide useful insights into the implications of the model parameters on the overall system performance. © 2011 IEEE.